ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libecb/ecb.h
Revision: 1.204
Committed: Fri Mar 25 08:44:14 2022 UTC (2 years, 2 months ago) by root
Content type: text/plain
Branch: MAIN
Changes since 1.203: +149 -1 lines
Log Message:
add hilbert2d

File Contents

# User Rev Content
1 root 1.1 /*
2 root 1.17 * libecb - http://software.schmorp.de/pkg/libecb
3 root 1.1 *
4 root 1.189 * Copyright (©) 2009-2015,2018-2021 Marc Alexander Lehmann <libecb@schmorp.de>
5 root 1.7 * Copyright (©) 2011 Emanuele Giaquinta
6 root 1.1 * All rights reserved.
7     *
8     * Redistribution and use in source and binary forms, with or without modifica-
9     * tion, are permitted provided that the following conditions are met:
10     *
11     * 1. Redistributions of source code must retain the above copyright notice,
12     * this list of conditions and the following disclaimer.
13     *
14     * 2. Redistributions in binary form must reproduce the above copyright
15     * notice, this list of conditions and the following disclaimer in the
16     * documentation and/or other materials provided with the distribution.
17     *
18     * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
19     * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
20     * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
21     * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
22     * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
23     * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
24     * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
25     * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
26     * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
27     * OF THE POSSIBILITY OF SUCH DAMAGE.
28 root 1.133 *
29     * Alternatively, the contents of this file may be used under the terms of
30     * the GNU General Public License ("GPL") version 2 or any later version,
31     * in which case the provisions of the GPL are applicable instead of
32     * the above. If you wish to allow the use of your version of this file
33     * only under the terms of the GPL and not to allow others to use your
34     * version of this file under the BSD license, indicate your decision
35     * by deleting the provisions above and replace them with the notice
36     * and other provisions required by the GPL. If you do not delete the
37     * provisions above, a recipient may use your version of this file under
38     * either the BSD or the GPL.
39 root 1.1 */
40    
41     #ifndef ECB_H
42     #define ECB_H
43    
44 root 1.87 /* 16 bits major, 16 bits minor */
45 root 1.204 #define ECB_VERSION 0x0001000c
46 root 1.87
47 root 1.184 #include <string.h> /* for memcpy */
48    
49 root 1.187 #if defined (_WIN32) && !defined (__MINGW32__)
50 root 1.44 typedef signed char int8_t;
51     typedef unsigned char uint8_t;
52 root 1.180 typedef signed char int_fast8_t;
53     typedef unsigned char uint_fast8_t;
54 root 1.44 typedef signed short int16_t;
55     typedef unsigned short uint16_t;
56 root 1.180 typedef signed int int_fast16_t;
57     typedef unsigned int uint_fast16_t;
58 root 1.44 typedef signed int int32_t;
59     typedef unsigned int uint32_t;
60 root 1.180 typedef signed int int_fast32_t;
61     typedef unsigned int uint_fast32_t;
62 root 1.44 #if __GNUC__
63     typedef signed long long int64_t;
64     typedef unsigned long long uint64_t;
65 root 1.51 #else /* _MSC_VER || __BORLANDC__ */
66 root 1.44 typedef signed __int64 int64_t;
67     typedef unsigned __int64 uint64_t;
68     #endif
69 root 1.180 typedef int64_t int_fast64_t;
70     typedef uint64_t uint_fast64_t;
71 root 1.87 #ifdef _WIN64
72     #define ECB_PTRSIZE 8
73     typedef uint64_t uintptr_t;
74     typedef int64_t intptr_t;
75     #else
76     #define ECB_PTRSIZE 4
77     typedef uint32_t uintptr_t;
78     typedef int32_t intptr_t;
79     #endif
80 root 1.44 #else
81     #include <inttypes.h>
82 root 1.173 #if (defined INTPTR_MAX ? INTPTR_MAX : ULONG_MAX) > 0xffffffffU
83 root 1.87 #define ECB_PTRSIZE 8
84     #else
85     #define ECB_PTRSIZE 4
86     #endif
87 root 1.44 #endif
88 root 1.6
89 sf-exg 1.159 #define ECB_GCC_AMD64 (__amd64 || __amd64__ || __x86_64 || __x86_64__)
90     #define ECB_MSVC_AMD64 (_M_AMD64 || _M_X64)
91    
92 root 1.179 #ifndef ECB_OPTIMIZE_SIZE
93     #if __OPTIMIZE_SIZE__
94     #define ECB_OPTIMIZE_SIZE 1
95     #else
96     #define ECB_OPTIMIZE_SIZE 0
97     #endif
98     #endif
99    
100 root 1.114 /* work around x32 idiocy by defining proper macros */
101 sf-exg 1.159 #if ECB_GCC_AMD64 || ECB_MSVC_AMD64
102 root 1.119 #if _ILP32
103 root 1.115 #define ECB_AMD64_X32 1
104 root 1.114 #else
105 root 1.115 #define ECB_AMD64 1
106 root 1.114 #endif
107     #endif
108    
109 root 1.189 #if ECB_PTRSIZE >= 8 || ECB_AMD64_X32
110     #define ECB_64BIT_NATIVE 1
111     #else
112     #define ECB_64BIT_NATIVE 0
113     #endif
114    
115 root 1.12 /* many compilers define _GNUC_ to some versions but then only implement
116     * what their idiot authors think are the "more important" extensions,
117 sf-exg 1.59 * causing enormous grief in return for some better fake benchmark numbers.
118 root 1.18 * or so.
119 root 1.12 * we try to detect these and simply assume they are not gcc - if they have
120     * an issue with that they should have done it right in the first place.
121     */
122 root 1.137 #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__
123     #define ECB_GCC_VERSION(major,minor) 0
124     #else
125     #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor)))
126 root 1.12 #endif
127 root 1.1
128 sf-exg 1.138 #define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor)))
129    
130 root 1.147 #if __clang__ && defined __has_builtin
131     #define ECB_CLANG_BUILTIN(x) __has_builtin (x)
132 sf-exg 1.138 #else
133     #define ECB_CLANG_BUILTIN(x) 0
134     #endif
135    
136 root 1.147 #if __clang__ && defined __has_extension
137     #define ECB_CLANG_EXTENSION(x) __has_extension (x)
138 sf-exg 1.140 #else
139     #define ECB_CLANG_EXTENSION(x) 0
140     #endif
141    
142 root 1.91 #define ECB_CPP (__cplusplus+0)
143     #define ECB_CPP11 (__cplusplus >= 201103L)
144 root 1.177 #define ECB_CPP14 (__cplusplus >= 201402L)
145     #define ECB_CPP17 (__cplusplus >= 201703L)
146 root 1.90
147 root 1.102 #if ECB_CPP
148 root 1.127 #define ECB_C 0
149     #define ECB_STDC_VERSION 0
150     #else
151     #define ECB_C 1
152     #define ECB_STDC_VERSION __STDC_VERSION__
153     #endif
154    
155     #define ECB_C99 (ECB_STDC_VERSION >= 199901L)
156     #define ECB_C11 (ECB_STDC_VERSION >= 201112L)
157 root 1.177 #define ECB_C17 (ECB_STDC_VERSION >= 201710L)
158 root 1.127
159     #if ECB_CPP
160 root 1.102 #define ECB_EXTERN_C extern "C"
161     #define ECB_EXTERN_C_BEG ECB_EXTERN_C {
162     #define ECB_EXTERN_C_END }
163     #else
164     #define ECB_EXTERN_C extern
165     #define ECB_EXTERN_C_BEG
166     #define ECB_EXTERN_C_END
167     #endif
168    
169 root 1.52 /*****************************************************************************/
170    
171 root 1.58 /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */
172     /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */
173    
174 root 1.79 #if ECB_NO_THREADS
175 root 1.95 #define ECB_NO_SMP 1
176 root 1.79 #endif
177    
178 root 1.93 #if ECB_NO_SMP
179 root 1.64 #define ECB_MEMORY_FENCE do { } while (0)
180 root 1.58 #endif
181    
182 sf-exg 1.165 /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */
183     #if __xlC__ && ECB_CPP
184     #include <builtins.h>
185     #endif
186    
187 root 1.171 #if 1400 <= _MSC_VER
188     #include <intrin.h> /* fence functions _ReadBarrier, also bit search functions _BitScanReverse */
189     #endif
190    
191 root 1.52 #ifndef ECB_MEMORY_FENCE
192 root 1.85 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
193 root 1.178 #define ECB_MEMORY_FENCE_RELAXED __asm__ __volatile__ ("" : : : "memory")
194 root 1.73 #if __i386 || __i386__
195 root 1.54 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
196 root 1.94 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
197 root 1.176 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
198 sf-exg 1.159 #elif ECB_GCC_AMD64
199 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
200     #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
201 root 1.176 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
202 root 1.63 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
203 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
204 root 1.175 #elif defined __ARM_ARCH_2__ \
205     || defined __ARM_ARCH_3__ || defined __ARM_ARCH_3M__ \
206     || defined __ARM_ARCH_4__ || defined __ARM_ARCH_4T__ \
207     || defined __ARM_ARCH_5__ || defined __ARM_ARCH_5E__ \
208     || defined __ARM_ARCH_5T__ || defined __ARM_ARCH_5TE__ \
209     || defined __ARM_ARCH_5TEJ__
210     /* should not need any, unless running old code on newer cpu - arm doesn't support that */
211 root 1.85 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
212 root 1.175 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \
213     || defined __ARM_ARCH_6T2__
214 root 1.84 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
215 root 1.85 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
216 root 1.175 || defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__
217 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
218 root 1.129 #elif __aarch64__
219     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
220 root 1.166 #elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8)
221 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
222     #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
223     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
224 root 1.85 #elif defined __s390__ || defined __s390x__
225 root 1.77 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
226 root 1.85 #elif defined __mips__
227 root 1.118 /* GNU/Linux emulates sync on mips1 architectures, so we force its use */
228 root 1.116 /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */
229     #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory")
230 root 1.86 #elif defined __alpha__
231 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")
232     #elif defined __hppa__
233     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
234     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
235     #elif defined __ia64__
236     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory")
237 root 1.117 #elif defined __m68k__
238     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
239     #elif defined __m88k__
240     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory")
241     #elif defined __sh__
242     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
243 root 1.52 #endif
244     #endif
245     #endif
246    
247     #ifndef ECB_MEMORY_FENCE
248 root 1.93 #if ECB_GCC_VERSION(4,7)
249 root 1.97 /* see comment below (stdatomic.h) about the C11 memory model. */
250 root 1.93 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
251 root 1.128 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
252     #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE)
253 root 1.190 #undef ECB_MEMORY_FENCE_RELAXED
254 root 1.178 #define ECB_MEMORY_FENCE_RELAXED __atomic_thread_fence (__ATOMIC_RELAXED)
255 root 1.110
256 sf-exg 1.140 #elif ECB_CLANG_EXTENSION(c_atomic)
257     /* see comment below (stdatomic.h) about the C11 memory model. */
258     #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
259     #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)
260     #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE)
261 root 1.190 #undef ECB_MEMORY_FENCE_RELAXED
262 root 1.178 #define ECB_MEMORY_FENCE_RELAXED __c11_atomic_thread_fence (__ATOMIC_RELAXED)
263 root 1.110
264 root 1.93 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
265 root 1.52 #define ECB_MEMORY_FENCE __sync_synchronize ()
266 root 1.126 #elif _MSC_VER >= 1500 /* VC++ 2008 */
267     /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */
268     #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
269     #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier()
270     #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */
271     #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier()
272 root 1.57 #elif _MSC_VER >= 1400 /* VC++ 2005 */
273     #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
274     #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
275     #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
276     #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
277 root 1.85 #elif defined _WIN32
278 root 1.55 #include <WinNT.h>
279 root 1.57 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
280 root 1.72 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
281     #include <mbarrier.h>
282 root 1.178 #define ECB_MEMORY_FENCE __machine_rw_barrier ()
283     #define ECB_MEMORY_FENCE_ACQUIRE __machine_acq_barrier ()
284     #define ECB_MEMORY_FENCE_RELEASE __machine_rel_barrier ()
285     #define ECB_MEMORY_FENCE_RELAXED __compiler_barrier ()
286 root 1.82 #elif __xlC__
287 root 1.83 #define ECB_MEMORY_FENCE __sync ()
288 root 1.52 #endif
289     #endif
290    
291 root 1.53 #ifndef ECB_MEMORY_FENCE
292 root 1.94 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
293     /* we assume that these memory fences work on all variables/all memory accesses, */
294     /* not just C11 atomics and atomic accesses */
295     #include <stdatomic.h>
296     #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
297 root 1.178 #define ECB_MEMORY_FENCE_ACQUIRE atomic_thread_fence (memory_order_acquire)
298     #define ECB_MEMORY_FENCE_RELEASE atomic_thread_fence (memory_order_release)
299 root 1.94 #endif
300     #endif
301    
302     #ifndef ECB_MEMORY_FENCE
303 root 1.62 #if !ECB_AVOID_PTHREADS
304     /*
305     * if you get undefined symbol references to pthread_mutex_lock,
306     * or failure to find pthread.h, then you should implement
307     * the ECB_MEMORY_FENCE operations for your cpu/compiler
308     * OR provide pthread.h and link against the posix thread library
309     * of your system.
310     */
311     #include <pthread.h>
312     #define ECB_NEEDS_PTHREADS 1
313     #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1
314 root 1.52
315 root 1.62 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
316     #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
317     #endif
318     #endif
319    
320 root 1.85 #if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE
321 root 1.52 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
322 root 1.62 #endif
323    
324 root 1.85 #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
325 root 1.52 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
326     #endif
327    
328 root 1.178 #if !defined ECB_MEMORY_FENCE_RELAXED && defined ECB_MEMORY_FENCE
329     #define ECB_MEMORY_FENCE_RELAXED ECB_MEMORY_FENCE /* very heavy-handed */
330     #endif
331    
332 root 1.52 /*****************************************************************************/
333    
334 root 1.149 #if ECB_CPP
335 root 1.46 #define ecb_inline static inline
336 root 1.38 #elif ECB_GCC_VERSION(2,5)
337 root 1.46 #define ecb_inline static __inline__
338 root 1.39 #elif ECB_C99
339 root 1.46 #define ecb_inline static inline
340 root 1.29 #else
341 root 1.46 #define ecb_inline static
342 root 1.38 #endif
343    
344     #if ECB_GCC_VERSION(3,3)
345     #define ecb_restrict __restrict__
346 root 1.39 #elif ECB_C99
347 root 1.38 #define ecb_restrict restrict
348     #else
349     #define ecb_restrict
350 root 1.4 #endif
351    
352 root 1.38 typedef int ecb_bool;
353    
354 root 1.8 #define ECB_CONCAT_(a, b) a ## b
355     #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
356     #define ECB_STRINGIFY_(a) # a
357     #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
358 root 1.155 #define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr))
359 root 1.8
360 root 1.46 #define ecb_function_ ecb_inline
361 root 1.3
362 sf-exg 1.138 #if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)
363 root 1.142 #define ecb_attribute(attrlist) __attribute__ (attrlist)
364 root 1.37 #else
365     #define ecb_attribute(attrlist)
366 sf-exg 1.138 #endif
367 root 1.127
368 sf-exg 1.138 #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p)
369     #define ecb_is_constant(expr) __builtin_constant_p (expr)
370     #else
371 root 1.127 /* possible C11 impl for integral types
372     typedef struct ecb_is_constant_struct ecb_is_constant_struct;
373     #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */
374    
375 root 1.37 #define ecb_is_constant(expr) 0
376 sf-exg 1.138 #endif
377    
378     #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect)
379     #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
380     #else
381 root 1.37 #define ecb_expect(expr,value) (expr)
382 sf-exg 1.138 #endif
383    
384     #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch)
385     #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
386     #else
387 root 1.37 #define ecb_prefetch(addr,rw,locality)
388 root 1.1 #endif
389    
390 root 1.2 /* no emulation for ecb_decltype */
391 root 1.143 #if ECB_CPP11
392 root 1.144 // older implementations might have problems with decltype(x)::type, work around it
393 root 1.146 template<class T> struct ecb_decltype_t { typedef T type; };
394     #define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type
395 root 1.143 #elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8)
396     #define ecb_decltype(x) __typeof__ (x)
397 root 1.1 #endif
398    
399 root 1.135 #if _MSC_VER >= 1300
400 root 1.149 #define ecb_deprecated __declspec (deprecated)
401 root 1.135 #else
402     #define ecb_deprecated ecb_attribute ((__deprecated__))
403     #endif
404    
405 sf-exg 1.162 #if _MSC_VER >= 1500
406 root 1.154 #define ecb_deprecated_message(msg) __declspec (deprecated (msg))
407     #elif ECB_GCC_VERSION(4,5)
408     #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg))
409     #else
410     #define ecb_deprecated_message(msg) ecb_deprecated
411     #endif
412    
413     #if _MSC_VER >= 1400
414     #define ecb_noinline __declspec (noinline)
415     #else
416     #define ecb_noinline ecb_attribute ((__noinline__))
417     #endif
418    
419 root 1.24 #define ecb_unused ecb_attribute ((__unused__))
420     #define ecb_const ecb_attribute ((__const__))
421     #define ecb_pure ecb_attribute ((__pure__))
422 root 1.35
423 root 1.145 #if ECB_C11 || __IBMC_NORETURN
424 sf-exg 1.165 /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */
425 root 1.90 #define ecb_noreturn _Noreturn
426 root 1.153 #elif ECB_CPP11
427     #define ecb_noreturn [[noreturn]]
428     #elif _MSC_VER >= 1200
429 sf-exg 1.156 /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */
430 root 1.153 #define ecb_noreturn __declspec (noreturn)
431 root 1.90 #else
432     #define ecb_noreturn ecb_attribute ((__noreturn__))
433     #endif
434    
435 root 1.35 #if ECB_GCC_VERSION(4,3)
436 root 1.39 #define ecb_artificial ecb_attribute ((__artificial__))
437     #define ecb_hot ecb_attribute ((__hot__))
438     #define ecb_cold ecb_attribute ((__cold__))
439 root 1.35 #else
440     #define ecb_artificial
441     #define ecb_hot
442     #define ecb_cold
443     #endif
444 root 1.1
445 root 1.39 /* put around conditional expressions if you are very sure that the */
446     /* expression is mostly true or mostly false. note that these return */
447     /* booleans, not the expression. */
448 root 1.33 #define ecb_expect_false(expr) ecb_expect (!!(expr), 0)
449     #define ecb_expect_true(expr) ecb_expect (!!(expr), 1)
450 root 1.36 /* for compatibility to the rest of the world */
451 root 1.33 #define ecb_likely(expr) ecb_expect_true (expr)
452     #define ecb_unlikely(expr) ecb_expect_false (expr)
453 root 1.1
454 root 1.3 /* count trailing zero bits and count # of one bits */
455 root 1.139 #if ECB_GCC_VERSION(3,4) \
456     || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \
457     && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \
458     && ECB_CLANG_BUILTIN(__builtin_popcount))
459 root 1.49 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
460     #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
461     #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
462 root 1.35 #define ecb_ctz32(x) __builtin_ctz (x)
463 root 1.49 #define ecb_ctz64(x) __builtin_ctzll (x)
464 root 1.35 #define ecb_popcount32(x) __builtin_popcount (x)
465 root 1.49 /* no popcountll */
466 root 1.1 #else
467 root 1.151 ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
468     ecb_function_ ecb_const int
469 root 1.35 ecb_ctz32 (uint32_t x)
470     {
471 root 1.171 #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
472 root 1.172 unsigned long r;
473 root 1.171 _BitScanForward (&r, x);
474     return (int)r;
475     #else
476 root 1.35 int r = 0;
477    
478 root 1.48 x &= ~x + 1; /* this isolates the lowest bit */
479 root 1.35
480 root 1.50 #if ECB_branchless_on_i386
481     r += !!(x & 0xaaaaaaaa) << 0;
482     r += !!(x & 0xcccccccc) << 1;
483     r += !!(x & 0xf0f0f0f0) << 2;
484     r += !!(x & 0xff00ff00) << 3;
485     r += !!(x & 0xffff0000) << 4;
486     #else
487 root 1.35 if (x & 0xaaaaaaaa) r += 1;
488     if (x & 0xcccccccc) r += 2;
489     if (x & 0xf0f0f0f0) r += 4;
490     if (x & 0xff00ff00) r += 8;
491     if (x & 0xffff0000) r += 16;
492 root 1.50 #endif
493 root 1.35
494     return r;
495 root 1.171 #endif
496 root 1.35 }
497    
498 root 1.151 ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
499     ecb_function_ ecb_const int
500 root 1.49 ecb_ctz64 (uint64_t x)
501     {
502 root 1.171 #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
503 root 1.172 unsigned long r;
504 root 1.171 _BitScanForward64 (&r, x);
505     return (int)r;
506     #else
507 root 1.168 int shift = x & 0xffffffff ? 0 : 32;
508 root 1.50 return ecb_ctz32 (x >> shift) + shift;
509 root 1.171 #endif
510 root 1.49 }
511    
512 root 1.151 ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
513     ecb_function_ ecb_const int
514 root 1.35 ecb_popcount32 (uint32_t x)
515     {
516     x -= (x >> 1) & 0x55555555;
517     x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
518     x = ((x >> 4) + x) & 0x0f0f0f0f;
519     x *= 0x01010101;
520 root 1.1
521 root 1.35 return x >> 24;
522     }
523 root 1.49
524 root 1.151 ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
525     ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
526 root 1.49 {
527 root 1.171 #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
528 root 1.172 unsigned long r;
529 root 1.171 _BitScanReverse (&r, x);
530     return (int)r;
531     #else
532 root 1.50 int r = 0;
533 root 1.49
534 root 1.50 if (x >> 16) { x >>= 16; r += 16; }
535     if (x >> 8) { x >>= 8; r += 8; }
536     if (x >> 4) { x >>= 4; r += 4; }
537     if (x >> 2) { x >>= 2; r += 2; }
538     if (x >> 1) { r += 1; }
539 root 1.49
540     return r;
541 root 1.171 #endif
542 root 1.49 }
543    
544 root 1.151 ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
545     ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
546 root 1.49 {
547 root 1.171 #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
548 root 1.172 unsigned long r;
549 root 1.171 _BitScanReverse64 (&r, x);
550     return (int)r;
551     #else
552 root 1.50 int r = 0;
553 root 1.49
554 root 1.50 if (x >> 32) { x >>= 32; r += 32; }
555 root 1.49
556 root 1.50 return r + ecb_ld32 (x);
557 root 1.171 #endif
558 root 1.49 }
559 root 1.1 #endif
560    
561 root 1.151 ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);
562     ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
563     ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x);
564     ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
565 root 1.88
566 root 1.151 ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x);
567     ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x)
568 root 1.70 {
569     return ( (x * 0x0802U & 0x22110U)
570 root 1.151 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
571 root 1.70 }
572    
573 root 1.151 ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x);
574     ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x)
575 root 1.70 {
576     x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
577     x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
578     x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
579     x = ( x >> 8 ) | ( x << 8);
580    
581     return x;
582     }
583    
584 root 1.151 ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x);
585     ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x)
586 root 1.70 {
587     x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
588     x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
589     x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
590     x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
591     x = ( x >> 16 ) | ( x << 16);
592    
593     return x;
594     }
595    
596 root 1.49 /* popcount64 is only available on 64 bit cpus as gcc builtin */
597     /* so for this version we are lazy */
598 root 1.151 ecb_function_ ecb_const int ecb_popcount64 (uint64_t x);
599     ecb_function_ ecb_const int
600 root 1.49 ecb_popcount64 (uint64_t x)
601     {
602     return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
603     }
604    
605 root 1.151 ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count);
606     ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count);
607     ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count);
608     ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count);
609     ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count);
610     ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count);
611     ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count);
612     ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count);
613    
614 root 1.198 ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> (-count & 7)) | (x << (count & 7)); }
615     ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << (-count & 7)) | (x >> (count & 7)); }
616     ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (-count & 15)) | (x << (count & 15)); }
617     ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (-count & 15)) | (x >> (count & 15)); }
618     ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (-count & 31)) | (x << (count & 31)); }
619     ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (-count & 31)) | (x >> (count & 31)); }
620     ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (-count & 63)) | (x << (count & 63)); }
621     ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (-count & 63)) | (x >> (count & 63)); }
622 root 1.50
623 root 1.182 #if ECB_CPP
624    
625     inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); }
626     inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); }
627     inline uint32_t ecb_ctz (uint32_t v) { return ecb_ctz32 (v); }
628     inline uint64_t ecb_ctz (uint64_t v) { return ecb_ctz64 (v); }
629    
630     inline bool ecb_is_pot (uint8_t v) { return ecb_is_pot32 (v); }
631     inline bool ecb_is_pot (uint16_t v) { return ecb_is_pot32 (v); }
632     inline bool ecb_is_pot (uint32_t v) { return ecb_is_pot32 (v); }
633     inline bool ecb_is_pot (uint64_t v) { return ecb_is_pot64 (v); }
634    
635     inline int ecb_ld (uint8_t v) { return ecb_ld32 (v); }
636     inline int ecb_ld (uint16_t v) { return ecb_ld32 (v); }
637     inline int ecb_ld (uint32_t v) { return ecb_ld32 (v); }
638     inline int ecb_ld (uint64_t v) { return ecb_ld64 (v); }
639    
640     inline int ecb_popcount (uint8_t v) { return ecb_popcount32 (v); }
641     inline int ecb_popcount (uint16_t v) { return ecb_popcount32 (v); }
642     inline int ecb_popcount (uint32_t v) { return ecb_popcount32 (v); }
643     inline int ecb_popcount (uint64_t v) { return ecb_popcount64 (v); }
644    
645     inline uint8_t ecb_bitrev (uint8_t v) { return ecb_bitrev8 (v); }
646     inline uint16_t ecb_bitrev (uint16_t v) { return ecb_bitrev16 (v); }
647     inline uint32_t ecb_bitrev (uint32_t v) { return ecb_bitrev32 (v); }
648    
649 root 1.183 inline uint8_t ecb_rotl (uint8_t v, unsigned int count) { return ecb_rotl8 (v, count); }
650     inline uint16_t ecb_rotl (uint16_t v, unsigned int count) { return ecb_rotl16 (v, count); }
651     inline uint32_t ecb_rotl (uint32_t v, unsigned int count) { return ecb_rotl32 (v, count); }
652     inline uint64_t ecb_rotl (uint64_t v, unsigned int count) { return ecb_rotl64 (v, count); }
653    
654     inline uint8_t ecb_rotr (uint8_t v, unsigned int count) { return ecb_rotr8 (v, count); }
655     inline uint16_t ecb_rotr (uint16_t v, unsigned int count) { return ecb_rotr16 (v, count); }
656     inline uint32_t ecb_rotr (uint32_t v, unsigned int count) { return ecb_rotr32 (v, count); }
657     inline uint64_t ecb_rotr (uint64_t v, unsigned int count) { return ecb_rotr64 (v, count); }
658 root 1.182
659     #endif
660    
661 sf-exg 1.138 #if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
662 root 1.164 #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16)
663     #define ecb_bswap16(x) __builtin_bswap16 (x)
664     #else
665 root 1.49 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
666 root 1.164 #endif
667 root 1.49 #define ecb_bswap32(x) __builtin_bswap32 (x)
668     #define ecb_bswap64(x) __builtin_bswap64 (x)
669 root 1.164 #elif _MSC_VER
670     #include <stdlib.h>
671     #define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x)))
672     #define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x)))
673     #define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x)))
674 root 1.13 #else
675 root 1.151 ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x);
676     ecb_function_ ecb_const uint16_t
677 root 1.50 ecb_bswap16 (uint16_t x)
678 root 1.49 {
679 root 1.50 return ecb_rotl16 (x, 8);
680 root 1.49 }
681    
682 root 1.151 ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x);
683     ecb_function_ ecb_const uint32_t
684 root 1.35 ecb_bswap32 (uint32_t x)
685     {
686 root 1.50 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
687 root 1.35 }
688    
689 root 1.151 ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x);
690     ecb_function_ ecb_const uint64_t
691 root 1.49 ecb_bswap64 (uint64_t x)
692 root 1.35 {
693 root 1.50 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
694 root 1.35 }
695 root 1.13 #endif
696    
697 sf-exg 1.138 #if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)
698 root 1.35 #define ecb_unreachable() __builtin_unreachable ()
699 root 1.13 #else
700 root 1.35 /* this seems to work fine, but gcc always emits a warning for it :/ */
701 root 1.151 ecb_inline ecb_noreturn void ecb_unreachable (void);
702     ecb_inline ecb_noreturn void ecb_unreachable (void) { }
703 root 1.13 #endif
704    
705 root 1.41 /* try to tell the compiler that some condition is definitely true */
706 root 1.100 #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
707 root 1.41
708 root 1.174 ecb_inline ecb_const uint32_t ecb_byteorder_helper (void);
709     ecb_inline ecb_const uint32_t
710 root 1.23 ecb_byteorder_helper (void)
711 root 1.3 {
712 root 1.98 /* the union code still generates code under pressure in gcc, */
713 sf-exg 1.111 /* but less than using pointers, and always seems to */
714 root 1.98 /* successfully return a constant. */
715     /* the reason why we have this horrible preprocessor mess */
716 root 1.99 /* is to avoid it in all cases, at least on common architectures */
717 sf-exg 1.111 /* or when using a recent enough gcc version (>= 4.6) */
718 root 1.174 #if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
719     || ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__)
720     #define ECB_LITTLE_ENDIAN 1
721     return 0x44332211;
722     #elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \
723     || ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__)
724     #define ECB_BIG_ENDIAN 1
725     return 0x11223344;
726 root 1.98 #else
727     union
728     {
729 root 1.174 uint8_t c[4];
730     uint32_t u;
731     } u = { 0x11, 0x22, 0x33, 0x44 };
732     return u.u;
733 root 1.98 #endif
734 root 1.3 }
735    
736 root 1.151 ecb_inline ecb_const ecb_bool ecb_big_endian (void);
737 root 1.174 ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; }
738 root 1.151 ecb_inline ecb_const ecb_bool ecb_little_endian (void);
739 root 1.174 ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; }
740 root 1.3
741 root 1.180 /*****************************************************************************/
742     /* unaligned load/store */
743    
744     ecb_inline uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
745     ecb_inline uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
746     ecb_inline uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
747    
748     ecb_inline uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
749     ecb_inline uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
750     ecb_inline uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
751    
752     ecb_inline uint_fast16_t ecb_peek_u16_u (const void *ptr) { uint16_t v; memcpy (&v, ptr, sizeof (v)); return v; }
753     ecb_inline uint_fast32_t ecb_peek_u32_u (const void *ptr) { uint32_t v; memcpy (&v, ptr, sizeof (v)); return v; }
754     ecb_inline uint_fast64_t ecb_peek_u64_u (const void *ptr) { uint64_t v; memcpy (&v, ptr, sizeof (v)); return v; }
755    
756     ecb_inline uint_fast16_t ecb_peek_be_u16_u (const void *ptr) { return ecb_be_u16_to_host (ecb_peek_u16_u (ptr)); }
757     ecb_inline uint_fast32_t ecb_peek_be_u32_u (const void *ptr) { return ecb_be_u32_to_host (ecb_peek_u32_u (ptr)); }
758     ecb_inline uint_fast64_t ecb_peek_be_u64_u (const void *ptr) { return ecb_be_u64_to_host (ecb_peek_u64_u (ptr)); }
759    
760     ecb_inline uint_fast16_t ecb_peek_le_u16_u (const void *ptr) { return ecb_le_u16_to_host (ecb_peek_u16_u (ptr)); }
761     ecb_inline uint_fast32_t ecb_peek_le_u32_u (const void *ptr) { return ecb_le_u32_to_host (ecb_peek_u32_u (ptr)); }
762     ecb_inline uint_fast64_t ecb_peek_le_u64_u (const void *ptr) { return ecb_le_u64_to_host (ecb_peek_u64_u (ptr)); }
763    
764     ecb_inline uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
765     ecb_inline uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
766     ecb_inline uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
767    
768     ecb_inline uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
769     ecb_inline uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
770     ecb_inline uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
771    
772     ecb_inline void ecb_poke_u16_u (void *ptr, uint16_t v) { memcpy (ptr, &v, sizeof (v)); }
773     ecb_inline void ecb_poke_u32_u (void *ptr, uint32_t v) { memcpy (ptr, &v, sizeof (v)); }
774     ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); }
775    
776     ecb_inline void ecb_poke_be_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_be_u16 (v)); }
777     ecb_inline void ecb_poke_be_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_be_u32 (v)); }
778     ecb_inline void ecb_poke_be_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_be_u64 (v)); }
779 sf-exg 1.196
780 root 1.180 ecb_inline void ecb_poke_le_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_le_u16 (v)); }
781     ecb_inline void ecb_poke_le_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_le_u32 (v)); }
782     ecb_inline void ecb_poke_le_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_le_u64 (v)); }
783    
784 root 1.186 #if ECB_CPP
785 root 1.180
786     inline uint8_t ecb_bswap (uint8_t v) { return v; }
787     inline uint16_t ecb_bswap (uint16_t v) { return ecb_bswap16 (v); }
788     inline uint32_t ecb_bswap (uint32_t v) { return ecb_bswap32 (v); }
789     inline uint64_t ecb_bswap (uint64_t v) { return ecb_bswap64 (v); }
790    
791     template<typename T> inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
792     template<typename T> inline T ecb_le_to_host (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
793     template<typename T> inline T ecb_peek (const void *ptr) { return *(const T *)ptr; }
794     template<typename T> inline T ecb_peek_be (const void *ptr) { return ecb_be_to_host (ecb_peek <T> (ptr)); }
795     template<typename T> inline T ecb_peek_le (const void *ptr) { return ecb_le_to_host (ecb_peek <T> (ptr)); }
796 root 1.184 template<typename T> inline T ecb_peek_u (const void *ptr) { T v; memcpy (&v, ptr, sizeof (v)); return v; }
797 root 1.180 template<typename T> inline T ecb_peek_be_u (const void *ptr) { return ecb_be_to_host (ecb_peek_u<T> (ptr)); }
798     template<typename T> inline T ecb_peek_le_u (const void *ptr) { return ecb_le_to_host (ecb_peek_u<T> (ptr)); }
799    
800     template<typename T> inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
801     template<typename T> inline T ecb_host_to_le (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
802     template<typename T> inline void ecb_poke (void *ptr, T v) { *(T *)ptr = v; }
803     template<typename T> inline void ecb_poke_be (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_be (v)); }
804     template<typename T> inline void ecb_poke_le (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_le (v)); }
805 root 1.184 template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); }
806 root 1.180 template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); }
807     template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); }
808    
809     #endif
810    
811     /*****************************************************************************/
812 root 1.199 /* pointer/integer hashing */
813    
814     /* based on hash by Chris Wellons, https://nullprogram.com/blog/2018/07/31/ */
815     ecb_function_ uint32_t ecb_mix32 (uint32_t v);
816     ecb_function_ uint32_t ecb_mix32 (uint32_t v)
817     {
818     v ^= v >> 16; v *= 0x7feb352dU;
819     v ^= v >> 15; v *= 0x846ca68bU;
820     v ^= v >> 16;
821     return v;
822     }
823    
824     ecb_function_ uint32_t ecb_unmix32 (uint32_t v);
825     ecb_function_ uint32_t ecb_unmix32 (uint32_t v)
826     {
827     v ^= v >> 16 ; v *= 0x43021123U;
828     v ^= v >> 15 ^ v >> 30; v *= 0x1d69e2a5U;
829     v ^= v >> 16 ;
830     return v;
831     }
832    
833     /* based on splitmix64, by Sebastiona Vigna, https://prng.di.unimi.it/splitmix64.c */
834     ecb_function_ uint64_t ecb_mix64 (uint64_t v);
835     ecb_function_ uint64_t ecb_mix64 (uint64_t v)
836     {
837     v ^= v >> 30; v *= 0xbf58476d1ce4e5b9U;
838     v ^= v >> 27; v *= 0x94d049bb133111ebU;
839     v ^= v >> 31;
840     return v;
841     }
842    
843     ecb_function_ uint64_t ecb_unmix64 (uint64_t v);
844     ecb_function_ uint64_t ecb_unmix64 (uint64_t v)
845     {
846     v ^= v >> 31 ^ v >> 62; v *= 0x319642b2d24d8ec3U;
847     v ^= v >> 27 ^ v >> 54; v *= 0x96de1b173f119089U;
848     v ^= v >> 30 ^ v >> 60;
849     return v;
850     }
851    
852     ecb_function_ uintptr_t ecb_ptrmix (void *p);
853     ecb_function_ uintptr_t ecb_ptrmix (void *p)
854     {
855     #if ECB_PTRSIZE <= 4
856     return ecb_mix32 ((uint32_t)p);
857     #else
858     return ecb_mix64 ((uint64_t)p);
859     #endif
860     }
861    
862     ecb_function_ void *ecb_ptrunmix (uintptr_t v);
863     ecb_function_ void *ecb_ptrunmix (uintptr_t v)
864     {
865     #if ECB_PTRSIZE <= 4
866     return (void *)ecb_unmix32 (v);
867     #else
868     return (void *)ecb_unmix64 (v);
869     #endif
870     }
871    
872     #if ECB_CPP
873    
874     template<typename T>
875     inline uintptr_t ecb_ptrmix (T *p)
876     {
877     return ecb_ptrmix (static_cast<void *>(p));
878     }
879    
880     template<typename T>
881     inline T *ecb_ptrunmix (uintptr_t v)
882     {
883     return static_cast<T *>(ecb_ptrunmix (v));
884     }
885    
886     #endif
887    
888     /*****************************************************************************/
889 root 1.202 /* gray code */
890    
891     ecb_function_ uint_fast8_t ecb_gray8_encode (uint_fast8_t b) { return b ^ (b >> 1); }
892     ecb_function_ uint_fast16_t ecb_gray16_encode (uint_fast16_t b) { return b ^ (b >> 1); }
893     ecb_function_ uint_fast32_t ecb_gray32_encode (uint_fast32_t b) { return b ^ (b >> 1); }
894     ecb_function_ uint_fast64_t ecb_gray64_encode (uint_fast64_t b) { return b ^ (b >> 1); }
895    
896     ecb_function_ uint8_t ecb_gray8_decode (uint8_t g)
897     {
898 root 1.203 g ^= g >> 1;
899     g ^= g >> 2;
900     g ^= g >> 4;
901    
902 root 1.202 return g;
903     }
904    
905     ecb_function_ uint16_t ecb_gray16_decode (uint16_t g)
906     {
907 root 1.203 g ^= g >> 1;
908     g ^= g >> 2;
909     g ^= g >> 4;
910     g ^= g >> 8;
911    
912 root 1.202 return g;
913     }
914    
915     ecb_function_ uint32_t ecb_gray32_decode (uint32_t g)
916     {
917 root 1.203 g ^= g >> 1;
918     g ^= g >> 2;
919     g ^= g >> 4;
920     g ^= g >> 8;
921     g ^= g >> 16;
922    
923 root 1.202 return g;
924     }
925    
926     ecb_function_ uint64_t ecb_gray64_decode (uint64_t g)
927     {
928 root 1.203 g ^= g >> 1;
929     g ^= g >> 2;
930     g ^= g >> 4;
931     g ^= g >> 8;
932     g ^= g >> 16;
933     g ^= g >> 32;
934    
935 root 1.202 return g;
936     }
937    
938     #if ECB_CPP
939    
940     ecb_function_ uint8_t ecb_gray_encode (uint8_t b) { return ecb_gray8_encode (b); }
941     ecb_function_ uint16_t ecb_gray_encode (uint16_t b) { return ecb_gray16_encode (b); }
942     ecb_function_ uint32_t ecb_gray_encode (uint32_t b) { return ecb_gray32_encode (b); }
943     ecb_function_ uint64_t ecb_gray_encode (uint64_t b) { return ecb_gray64_encode (b); }
944    
945     ecb_function_ uint8_t ecb_gray_decode (uint8_t g) { return ecb_gray8_decode (g); }
946     ecb_function_ uint16_t ecb_gray_decode (uint16_t g) { return ecb_gray16_decode (g); }
947     ecb_function_ uint32_t ecb_gray_decode (uint32_t g) { return ecb_gray32_decode (g); }
948     ecb_function_ uint64_t ecb_gray_decode (uint64_t g) { return ecb_gray64_decode (g); }
949    
950     #endif
951    
952     /*****************************************************************************/
953 root 1.204 /* 2d hilbert curves */
954    
955     /* algorithm from the book Hacker's Delight, modified to not */
956     /* run into undefined behaviour for n==16 */
957     static uint32_t
958     ecb_hilbert2d_index_to_coord32 (int n, uint32_t s)
959     {
960     uint32_t comp, swap, cs, t, sr;
961    
962     /* pad s on the left (unused) bits with 01 (no change groups) */
963     s |= 0x55555555U << n << n;
964     /* "s shift right" */
965     sr = (s >> 1) & 0x55555555U;
966     /* compute complement and swap info in two-bit groups */
967     cs = ((s & 0x55555555U) + sr) ^ 0x55555555U;
968    
969     /* parallel prefix xor op to propagate both complement
970     * and swap info together from left to right (there is
971     * no step "cs ^= cs >> 1", so in effect it computes
972     * two independent parallel prefix operations on two
973     * interleaved sets of sixteen bits).
974     */
975     cs ^= cs >> 2;
976     cs ^= cs >> 4;
977     cs ^= cs >> 8;
978     cs ^= cs >> 16;
979    
980     /* separate swap and complement bits */
981     swap = cs & 0x55555555U;
982     comp = (cs >> 1) & 0x55555555U;
983    
984     /* calculate coordinates in odd and even bit positions */
985     t = (s & swap) ^ comp;
986     s = s ^ sr ^ t ^ (t << 1);
987    
988     /* unpad/clear out any junk on the left */
989     s = s & ((1 << n << n) - 1);
990    
991     /* Now "unshuffle" to separate the x and y bits. */
992     t = (s ^ (s >> 1)) & 0x22222222U; s ^= t ^ (t << 1);
993     t = (s ^ (s >> 2)) & 0x0c0c0c0cU; s ^= t ^ (t << 2);
994     t = (s ^ (s >> 4)) & 0x00f000f0U; s ^= t ^ (t << 4);
995     t = (s ^ (s >> 8)) & 0x0000ff00U; s ^= t ^ (t << 8);
996    
997     /* now s contains two 16-bit coordinates */
998     return s;
999     }
1000    
1001     /* 64 bit, a straightforward extension to the 32 bit case */
1002     static uint64_t
1003     ecb_hilbert2d_index_to_coord64 (int n, uint64_t s)
1004     {
1005     uint64_t comp, swap, cs, t, sr;
1006    
1007     /* pad s on the left (unused) bits with 01 (no change groups) */
1008     s |= 0x5555555555555555U << n << n;
1009     /* "s shift right" */
1010     sr = (s >> 1) & 0x5555555555555555U;
1011     /* compute complement and swap info in two-bit groups */
1012     cs = ((s & 0x5555555555555555U) + sr) ^ 0x5555555555555555U;
1013    
1014     /* parallel prefix xor op to propagate both complement
1015     * and swap info together from left to right (there is
1016     * no step "cs ^= cs >> 1", so in effect it computes
1017     * two independent parallel prefix operations on two
1018     * interleaved sets of thirty-two bits).
1019     */
1020     cs ^= cs >> 2;
1021     cs ^= cs >> 4;
1022     cs ^= cs >> 8;
1023     cs ^= cs >> 16;
1024     cs ^= cs >> 32;
1025    
1026     /* separate swap and complement bits */
1027     swap = cs & 0x5555555555555555U;
1028     comp = (cs >> 1) & 0x5555555555555555U;
1029    
1030     /* calculate coordinates in odd and even bit positions */
1031     t = (s & swap) ^ comp;
1032     s = s ^ sr ^ t ^ (t << 1);
1033    
1034     /* unpad/clear out any junk on the left */
1035     s = s & ((1 << n << n) - 1);
1036    
1037     /* Now "unshuffle" to separate the x and y bits. */
1038     t = (s ^ (s >> 1)) & 0x2222222222222222U; s ^= t ^ (t << 1);
1039     t = (s ^ (s >> 2)) & 0x0c0c0c0c0c0c0c0cU; s ^= t ^ (t << 2);
1040     t = (s ^ (s >> 4)) & 0x00f000f000f000f0U; s ^= t ^ (t << 4);
1041     t = (s ^ (s >> 8)) & 0x0000ff000000ff00U; s ^= t ^ (t << 8);
1042     t = (s ^ (s >> 16)) & 0x00000000ffff0000U; s ^= t ^ (t << 16);
1043    
1044     /* now s contains two 32-bit coordinates */
1045     return s;
1046     }
1047    
1048     /* algorithm from the book Hacker's Delight, but a similar algorithm*/
1049     /* is given in https://doi.org/10.1002/spe.4380160103 */
1050     /* this has been slightly improved over the original version */
1051     ecb_function_ uint32_t
1052     ecb_hilbert2d_coord_to_index32 (int n, uint32_t xy)
1053     {
1054     uint32_t row;
1055     uint32_t state = 0;
1056     uint32_t s = 0;
1057    
1058     do
1059     {
1060     --n;
1061    
1062     row = 4 * state
1063     | (2 & (xy >> n >> 15))
1064     | (1 & (xy >> n ));
1065    
1066     /* these funky constants are lookup tables for two-bit values */
1067     s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3;
1068     state = (0x8fe65831U >> 2 * row) & 3;
1069     }
1070     while (n > 0);
1071    
1072     return s;
1073     }
1074    
1075     /* 64 bit, essentially the same as 32 bit */
1076     ecb_function_ uint64_t
1077     ecb_hilbert2d_coord_to_index64 (int n, uint64_t xy)
1078     {
1079     uint32_t row;
1080     uint32_t state = 0;
1081     uint64_t s = 0;
1082    
1083     do
1084     {
1085     --n;
1086    
1087     row = 4 * state
1088     | (2 & (xy >> n >> 31))
1089     | (1 & (xy >> n ));
1090    
1091     /* these funky constants are lookup tables for two-bit values */
1092     s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3;
1093     state = (0x8fe65831U >> 2 * row) & 3;
1094     }
1095     while (n > 0);
1096    
1097     return s;
1098     }
1099    
1100     /*****************************************************************************/
1101 root 1.188 /* division */
1102 root 1.180
1103 root 1.39 #if ECB_GCC_VERSION(3,0) || ECB_C99
1104 root 1.188 /* C99 tightened the definition of %, so we can use a more efficient version */
1105 root 1.35 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
1106 root 1.31 #else
1107 root 1.35 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
1108 root 1.31 #endif
1109 root 1.21
1110 root 1.149 #if ECB_CPP
1111 sf-exg 1.68 template<typename T>
1112     static inline T ecb_div_rd (T val, T div)
1113     {
1114     return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
1115     }
1116     template<typename T>
1117     static inline T ecb_div_ru (T val, T div)
1118     {
1119     return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
1120     }
1121     #else
1122     #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
1123     #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
1124     #endif
1125 sf-exg 1.67
1126 root 1.188 /*****************************************************************************/
1127     /* array length */
1128    
1129 root 1.5 #if ecb_cplusplus_does_not_suck
1130 root 1.40 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
1131 root 1.35 template<typename T, int N>
1132     static inline int ecb_array_length (const T (&arr)[N])
1133     {
1134     return N;
1135     }
1136 root 1.5 #else
1137 root 1.35 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
1138 root 1.5 #endif
1139    
1140 root 1.180 /*****************************************************************************/
1141 root 1.188 /* IEEE 754-2008 half float conversions */
1142 root 1.180
1143 root 1.170 ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x);
1144 root 1.167 ecb_function_ ecb_const uint32_t
1145 root 1.170 ecb_binary16_to_binary32 (uint32_t x)
1146 root 1.167 {
1147     unsigned int s = (x & 0x8000) << (31 - 15);
1148     int e = (x >> 10) & 0x001f;
1149     unsigned int m = x & 0x03ff;
1150    
1151     if (ecb_expect_false (e == 31))
1152     /* infinity or NaN */
1153     e = 255 - (127 - 15);
1154     else if (ecb_expect_false (!e))
1155     {
1156     if (ecb_expect_true (!m))
1157     /* zero, handled by code below by forcing e to 0 */
1158     e = 0 - (127 - 15);
1159     else
1160     {
1161     /* subnormal, renormalise */
1162     unsigned int s = 10 - ecb_ld32 (m);
1163    
1164     m = (m << s) & 0x3ff; /* mask implicit bit */
1165     e -= s - 1;
1166     }
1167     }
1168    
1169     /* e and m now are normalised, or zero, (or inf or nan) */
1170     e += 127 - 15;
1171    
1172     return s | (e << 23) | (m << (23 - 10));
1173     }
1174    
1175     ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x);
1176     ecb_function_ ecb_const uint16_t
1177     ecb_binary32_to_binary16 (uint32_t x)
1178     {
1179     unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */
1180 root 1.188 int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */
1181 root 1.167 unsigned int m = x & 0x007fffff;
1182    
1183     x &= 0x7fffffff;
1184    
1185     /* if it's within range of binary16 normals, use fast path */
1186     if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff))
1187     {
1188     /* mantissa round-to-even */
1189     m += 0x00000fff + ((m >> (23 - 10)) & 1);
1190    
1191     /* handle overflow */
1192     if (ecb_expect_false (m >= 0x00800000))
1193     {
1194     m >>= 1;
1195     e += 1;
1196     }
1197    
1198     return s | (e << 10) | (m >> (23 - 10));
1199     }
1200    
1201     /* handle large numbers and infinity */
1202     if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000))
1203     return s | 0x7c00;
1204    
1205 root 1.169 /* handle zero, subnormals and small numbers */
1206 root 1.167 if (ecb_expect_true (x < 0x38800000))
1207     {
1208     /* zero */
1209     if (ecb_expect_true (!x))
1210     return s;
1211    
1212     /* handle subnormals */
1213    
1214 root 1.169 /* too small, will be zero */
1215     if (e < (14 - 24)) /* might not be sharp, but is good enough */
1216     return s;
1217    
1218 root 1.167 m |= 0x00800000; /* make implicit bit explicit */
1219    
1220     /* very tricky - we need to round to the nearest e (+10) bit value */
1221     {
1222     unsigned int bits = 14 - e;
1223     unsigned int half = (1 << (bits - 1)) - 1;
1224     unsigned int even = (m >> bits) & 1;
1225    
1226     /* if this overflows, we will end up with a normalised number */
1227     m = (m + half + even) >> bits;
1228     }
1229    
1230     return s | m;
1231     }
1232    
1233     /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */
1234     m >>= 13;
1235    
1236     return s | 0x7c00 | m | !m;
1237     }
1238    
1239 root 1.104 /*******************************************************************************/
1240 root 1.191 /* fast integer to ascii */
1241    
1242 root 1.195 /*
1243     * This code is pretty complicated because it is general. The idea behind it,
1244     * however, is pretty simple: first, the number is multiplied with a scaling
1245 root 1.197 * factor (2**bits / 10**(digits-1)) to convert the integer into a fixed-point
1246 root 1.195 * number with the first digit in the upper bits.
1247     * Then this digit is converted to text and masked out. The resulting number
1248     * is then multiplied by 10, by multiplying the fixed point representation
1249     * by 5 and shifting the (binary) decimal point one to the right, so a 4.28
1250     * format becomes 5.27, 6.26 and so on.
1251     * The rest involves only advancing the pointer if we already generated a
1252     * non-zero digit, so leading zeroes are overwritten.
1253     */
1254    
1255 root 1.201 /* simply return a mask with "bits" bits set */
1256 root 1.191 #define ecb_i2a_mask(type,bits) ((((type)1) << (bits)) - 1)
1257    
1258 root 1.200 /* oputput a single digit. maskvalue is 10**digitidx */
1259 root 1.191 #define ecb_i2a_digit(type,bits,digitmask,maskvalue,digitidx) \
1260     if (digitmask >= maskvalue) /* constant, used to decide how many digits to generate */ \
1261     { \
1262     char digit = x >> (bits - digitidx); /* calculate the topmost digit */ \
1263     *ptr = digit + '0'; /* output it */ \
1264     nz = (digitmask == maskvalue) || nz || digit; /* first term == always output last digit */ \
1265     ptr += nz; /* output digit only if non-zero digit seen */ \
1266     x = (x & ecb_i2a_mask (type, bits - digitidx)) * 5; /* *10, but shift decimal point right */ \
1267     }
1268    
1269 root 1.200 /* convert integer to fixed point format and multiply out digits, highest first */
1270     /* requires magic constants: max. digits and number of bits after the decimal point */
1271 root 1.191 #define ecb_i2a_def(suffix,ptr,v,type,bits,digitmask,lz) \
1272     ecb_inline char *ecb_i2a_ ## suffix (char *ptr, uint32_t u) \
1273     { \
1274     char nz = lz; /* non-zero digit seen? */ \
1275     /* convert to x.bits fixed-point */ \
1276     type x = u * ((ecb_i2a_mask (type, bits) + digitmask) / digitmask); \
1277     /* output up to 10 digits */ \
1278     ecb_i2a_digit (type,bits,digitmask, 1, 0); \
1279     ecb_i2a_digit (type,bits,digitmask, 10, 1); \
1280     ecb_i2a_digit (type,bits,digitmask, 100, 2); \
1281     ecb_i2a_digit (type,bits,digitmask, 1000, 3); \
1282     ecb_i2a_digit (type,bits,digitmask, 10000, 4); \
1283     ecb_i2a_digit (type,bits,digitmask, 100000, 5); \
1284     ecb_i2a_digit (type,bits,digitmask, 1000000, 6); \
1285     ecb_i2a_digit (type,bits,digitmask, 10000000, 7); \
1286     ecb_i2a_digit (type,bits,digitmask, 100000000, 8); \
1287     ecb_i2a_digit (type,bits,digitmask, 1000000000, 9); \
1288     return ptr; \
1289     }
1290    
1291 root 1.200 /* predefined versions of the above, for various digits */
1292     /* ecb_i2a_xN = almost N digits, limit defined by macro */
1293     /* ecb_i2a_N = up to N digits, leading zeroes suppressed */
1294     /* ecb_i2a_0N = exactly N digits, including leading zeroes */
1295    
1296     /* non-leading-zero versions, limited range */
1297     #define ECB_I2A_MAX_X5 59074 /* limit for ecb_i2a_x5 */
1298     #define ECB_I2A_MAX_X10 2932500665 /* limit for ecb_i2a_x10 */
1299 root 1.191 ecb_i2a_def ( x5, ptr, v, uint32_t, 26, 10000, 0)
1300     ecb_i2a_def (x10, ptr, v, uint64_t, 60, 1000000000, 0)
1301    
1302 root 1.200 /* non-leading zero versions, all digits, 4 and 9 are optimal for 32/64 bit */
1303 root 1.194 ecb_i2a_def ( 2, ptr, v, uint32_t, 10, 10, 0)
1304     ecb_i2a_def ( 3, ptr, v, uint32_t, 12, 100, 0)
1305     ecb_i2a_def ( 4, ptr, v, uint32_t, 26, 1000, 0)
1306     ecb_i2a_def ( 5, ptr, v, uint64_t, 30, 10000, 0)
1307     ecb_i2a_def ( 6, ptr, v, uint64_t, 36, 100000, 0)
1308     ecb_i2a_def ( 7, ptr, v, uint64_t, 44, 1000000, 0)
1309     ecb_i2a_def ( 8, ptr, v, uint64_t, 50, 10000000, 0)
1310     ecb_i2a_def ( 9, ptr, v, uint64_t, 56, 100000000, 0)
1311 root 1.191
1312 root 1.200 /* leading-zero versions, all digits, 04 and 09 are optimal for 32/64 bit */
1313 root 1.194 ecb_i2a_def (02, ptr, v, uint32_t, 10, 10, 1)
1314     ecb_i2a_def (03, ptr, v, uint32_t, 12, 100, 1)
1315     ecb_i2a_def (04, ptr, v, uint32_t, 26, 1000, 1)
1316     ecb_i2a_def (05, ptr, v, uint64_t, 30, 10000, 1)
1317     ecb_i2a_def (06, ptr, v, uint64_t, 36, 100000, 1)
1318     ecb_i2a_def (07, ptr, v, uint64_t, 44, 1000000, 1)
1319     ecb_i2a_def (08, ptr, v, uint64_t, 50, 10000000, 1)
1320     ecb_i2a_def (09, ptr, v, uint64_t, 56, 100000000, 1)
1321 root 1.191
1322 root 1.192 #define ECB_I2A_I32_DIGITS 11
1323     #define ECB_I2A_U32_DIGITS 10
1324     #define ECB_I2A_I64_DIGITS 20
1325 root 1.194 #define ECB_I2A_U64_DIGITS 21
1326 root 1.193 #define ECB_I2A_MAX_DIGITS 21
1327 root 1.192
1328 root 1.191 ecb_inline char *
1329     ecb_i2a_u32 (char *ptr, uint32_t u)
1330     {
1331     #if ECB_64BIT_NATIVE
1332     if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1333     ptr = ecb_i2a_x10 (ptr, u);
1334 root 1.200 else /* x10 almost, but not fully, covers 32 bit */
1335 root 1.191 {
1336     uint32_t u1 = u % 1000000000;
1337     uint32_t u2 = u / 1000000000;
1338    
1339     *ptr++ = u2 + '0';
1340     ptr = ecb_i2a_09 (ptr, u1);
1341     }
1342     #else
1343     if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
1344     ecb_i2a_x5 (ptr, u);
1345     else if (ecb_expect_true (u <= ECB_I2A_MAX_X5 * 10000))
1346     {
1347     uint32_t u1 = u % 10000;
1348     uint32_t u2 = u / 10000;
1349    
1350     ptr = ecb_i2a_x5 (ptr, u2);
1351     ptr = ecb_i2a_04 (ptr, u1);
1352     }
1353     else
1354     {
1355     uint32_t u1 = u % 10000;
1356     uint32_t ua = u / 10000;
1357     uint32_t u2 = ua % 10000;
1358     uint32_t u3 = ua / 10000;
1359    
1360     ptr = ecb_i2a_2 (ptr, u3);
1361     ptr = ecb_i2a_04 (ptr, u2);
1362     ptr = ecb_i2a_04 (ptr, u1);
1363     }
1364     #endif
1365    
1366     return ptr;
1367     }
1368    
1369     ecb_inline char *
1370     ecb_i2a_i32 (char *ptr, int32_t v)
1371     {
1372     *ptr = '-'; ptr += v < 0;
1373     uint32_t u = v < 0 ? -(uint32_t)v : v;
1374    
1375     #if ECB_64BIT_NATIVE
1376 root 1.200 ptr = ecb_i2a_x10 (ptr, u); /* x10 fully covers 31 bit */
1377 root 1.191 #else
1378     ptr = ecb_i2a_u32 (ptr, u);
1379     #endif
1380    
1381     return ptr;
1382     }
1383    
1384     ecb_inline char *
1385     ecb_i2a_u64 (char *ptr, uint64_t u)
1386     {
1387     #if ECB_64BIT_NATIVE
1388     if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1389     ptr = ecb_i2a_x10 (ptr, u);
1390     else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
1391     {
1392     uint64_t u1 = u % 1000000000;
1393     uint64_t u2 = u / 1000000000;
1394    
1395     ptr = ecb_i2a_x10 (ptr, u2);
1396     ptr = ecb_i2a_09 (ptr, u1);
1397     }
1398     else
1399     {
1400     uint64_t u1 = u % 1000000000;
1401     uint64_t ua = u / 1000000000;
1402     uint64_t u2 = ua % 1000000000;
1403     uint64_t u3 = ua / 1000000000;
1404    
1405     ptr = ecb_i2a_2 (ptr, u3);
1406     ptr = ecb_i2a_09 (ptr, u2);
1407     ptr = ecb_i2a_09 (ptr, u1);
1408     }
1409     #else
1410     if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
1411     ptr = ecb_i2a_x5 (ptr, u);
1412     else
1413     {
1414     uint64_t u1 = u % 10000;
1415     uint64_t u2 = u / 10000;
1416    
1417     ptr = ecb_i2a_u64 (ptr, u2);
1418     ptr = ecb_i2a_04 (ptr, u1);
1419     }
1420     #endif
1421    
1422     return ptr;
1423     }
1424    
1425     ecb_inline char *
1426     ecb_i2a_i64 (char *ptr, int64_t v)
1427     {
1428     *ptr = '-'; ptr += v < 0;
1429     uint64_t u = v < 0 ? -(uint64_t)v : v;
1430    
1431     #if ECB_64BIT_NATIVE
1432     if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1433     ptr = ecb_i2a_x10 (ptr, u);
1434     else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
1435     {
1436     uint64_t u1 = u % 1000000000;
1437     uint64_t u2 = u / 1000000000;
1438    
1439     ptr = ecb_i2a_x10 (ptr, u2);
1440     ptr = ecb_i2a_09 (ptr, u1);
1441     }
1442     else
1443     {
1444     uint64_t u1 = u % 1000000000;
1445     uint64_t ua = u / 1000000000;
1446     uint64_t u2 = ua % 1000000000;
1447     uint64_t u3 = ua / 1000000000;
1448    
1449 root 1.200 /* 2**31 is 19 digits, so the top is exactly one digit */
1450 root 1.191 *ptr++ = u3 + '0';
1451     ptr = ecb_i2a_09 (ptr, u2);
1452     ptr = ecb_i2a_09 (ptr, u1);
1453     }
1454     #else
1455     ptr = ecb_i2a_u64 (ptr, u);
1456     #endif
1457    
1458     return ptr;
1459     }
1460    
1461     /*******************************************************************************/
1462 root 1.104 /* floating point stuff, can be disabled by defining ECB_NO_LIBM */
1463    
1464     /* basically, everything uses "ieee pure-endian" floating point numbers */
1465     /* the only noteworthy exception is ancient armle, which uses order 43218765 */
1466     #if 0 \
1467     || __i386 || __i386__ \
1468 sf-exg 1.159 || ECB_GCC_AMD64 \
1469 root 1.104 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
1470     || defined __s390__ || defined __s390x__ \
1471     || defined __mips__ \
1472     || defined __alpha__ \
1473     || defined __hppa__ \
1474     || defined __ia64__ \
1475 root 1.117 || defined __m68k__ \
1476     || defined __m88k__ \
1477     || defined __sh__ \
1478 sf-exg 1.159 || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \
1479 root 1.131 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
1480 root 1.132 || defined __aarch64__
1481 root 1.104 #define ECB_STDFP 1
1482 root 1.102 #else
1483 root 1.104 #define ECB_STDFP 0
1484 root 1.102 #endif
1485    
1486 root 1.104 #ifndef ECB_NO_LIBM
1487 root 1.103
1488 root 1.121 #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */
1489    
1490 root 1.122 /* only the oldest of old doesn't have this one. solaris. */
1491     #ifdef INFINITY
1492     #define ECB_INFINITY INFINITY
1493     #else
1494     #define ECB_INFINITY HUGE_VAL
1495     #endif
1496    
1497     #ifdef NAN
1498 root 1.121 #define ECB_NAN NAN
1499     #else
1500 root 1.122 #define ECB_NAN ECB_INFINITY
1501 root 1.121 #endif
1502 root 1.120
1503 root 1.148 #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L
1504 root 1.150 #define ecb_ldexpf(x,e) ldexpf ((x), (e))
1505 sf-exg 1.163 #define ecb_frexpf(x,e) frexpf ((x), (e))
1506 root 1.148 #else
1507 sf-exg 1.161 #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))
1508 sf-exg 1.163 #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))
1509 root 1.148 #endif
1510    
1511 root 1.104 /* convert a float to ieee single/binary32 */
1512 root 1.151 ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
1513     ecb_function_ ecb_const uint32_t
1514 root 1.103 ecb_float_to_binary32 (float x)
1515     {
1516     uint32_t r;
1517    
1518     #if ECB_STDFP
1519 root 1.104 memcpy (&r, &x, 4);
1520 root 1.103 #else
1521 root 1.105 /* slow emulation, works for anything but -0 */
1522 root 1.103 uint32_t m;
1523     int e;
1524    
1525 root 1.108 if (x == 0e0f ) return 0x00000000U;
1526 root 1.103 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
1527     if (x < -3.40282346638528860e+38f) return 0xff800000U;
1528 root 1.105 if (x != x ) return 0x7fbfffffU;
1529 root 1.103
1530 sf-exg 1.163 m = ecb_frexpf (x, &e) * 0x1000000U;
1531 root 1.103
1532     r = m & 0x80000000U;
1533    
1534     if (r)
1535     m = -m;
1536    
1537 root 1.108 if (e <= -126)
1538 root 1.103 {
1539     m &= 0xffffffU;
1540     m >>= (-125 - e);
1541     e = -126;
1542     }
1543    
1544     r |= (e + 126) << 23;
1545     r |= m & 0x7fffffU;
1546     #endif
1547    
1548     return r;
1549     }
1550    
1551 root 1.104 /* converts an ieee single/binary32 to a float */
1552 root 1.151 ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x);
1553     ecb_function_ ecb_const float
1554 root 1.103 ecb_binary32_to_float (uint32_t x)
1555     {
1556     float r;
1557    
1558     #if ECB_STDFP
1559 root 1.104 memcpy (&r, &x, 4);
1560 root 1.103 #else
1561     /* emulation, only works for normals and subnormals and +0 */
1562     int neg = x >> 31;
1563     int e = (x >> 23) & 0xffU;
1564    
1565     x &= 0x7fffffU;
1566    
1567     if (e)
1568     x |= 0x800000U;
1569 root 1.104 else
1570     e = 1;
1571 root 1.103
1572     /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
1573 root 1.148 r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126);
1574 root 1.103
1575     r = neg ? -r : r;
1576     #endif
1577    
1578     return r;
1579     }
1580    
1581 root 1.104 /* convert a double to ieee double/binary64 */
1582 root 1.151 ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x);
1583     ecb_function_ ecb_const uint64_t
1584 root 1.103 ecb_double_to_binary64 (double x)
1585     {
1586 root 1.104 uint64_t r;
1587    
1588     #if ECB_STDFP
1589     memcpy (&r, &x, 8);
1590     #else
1591 root 1.105 /* slow emulation, works for anything but -0 */
1592 root 1.104 uint64_t m;
1593     int e;
1594    
1595 root 1.108 if (x == 0e0 ) return 0x0000000000000000U;
1596 root 1.104 if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U;
1597     if (x < -1.79769313486231470e+308) return 0xfff0000000000000U;
1598 root 1.105 if (x != x ) return 0X7ff7ffffffffffffU;
1599 root 1.104
1600     m = frexp (x, &e) * 0x20000000000000U;
1601    
1602     r = m & 0x8000000000000000;;
1603    
1604     if (r)
1605     m = -m;
1606    
1607 root 1.108 if (e <= -1022)
1608 root 1.104 {
1609     m &= 0x1fffffffffffffU;
1610     m >>= (-1021 - e);
1611     e = -1022;
1612     }
1613    
1614     r |= ((uint64_t)(e + 1022)) << 52;
1615     r |= m & 0xfffffffffffffU;
1616     #endif
1617    
1618     return r;
1619     }
1620    
1621     /* converts an ieee double/binary64 to a double */
1622 root 1.151 ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x);
1623     ecb_function_ ecb_const double
1624 root 1.104 ecb_binary64_to_double (uint64_t x)
1625     {
1626     double r;
1627    
1628     #if ECB_STDFP
1629     memcpy (&r, &x, 8);
1630     #else
1631     /* emulation, only works for normals and subnormals and +0 */
1632     int neg = x >> 63;
1633     int e = (x >> 52) & 0x7ffU;
1634    
1635     x &= 0xfffffffffffffU;
1636    
1637     if (e)
1638     x |= 0x10000000000000U;
1639     else
1640     e = 1;
1641    
1642 root 1.107 /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */
1643 root 1.108 r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022);
1644 root 1.104
1645     r = neg ? -r : r;
1646     #endif
1647    
1648     return r;
1649 root 1.103 }
1650    
1651 root 1.167 /* convert a float to ieee half/binary16 */
1652     ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x);
1653     ecb_function_ ecb_const uint16_t
1654     ecb_float_to_binary16 (float x)
1655     {
1656     return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x));
1657     }
1658    
1659     /* convert an ieee half/binary16 to float */
1660     ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
1661     ecb_function_ ecb_const float
1662     ecb_binary16_to_float (uint16_t x)
1663     {
1664     return ecb_binary32_to_float (ecb_binary16_to_binary32 (x));
1665     }
1666    
1667 root 1.103 #endif
1668 root 1.102
1669 root 1.1 #endif
1670